Peter A. Jonsson

That sounds like a problem elsewhere. @emilaxelsson: is the `AlphaEq` instance for Conversion wrong in Feldspar? It uses `alphaEqSymDefault` which sounds perfectly sane.

The real problem is the equality test since two expressions of different types are considered equivalent despite the fact that we can observe they are different. Your explicit testing for...

For the books: workaround merged as #15. Remember to remove that code when this bug is fixed. Changing title of the ticket to reflect the underlying issue.

I've been waiting on getting the PR queue merged since that touches these areas, but it's possible stream-unzip (https://pypi.org/project/stream-unzip/) could be used to get the unpacked results saving one IO...

Edit: fixed my confusion about os.close. I don't have a reproducer, but I also see faulty zip files which I assume is caused by this issue. I'm using a wired...

I don't think my PR solves this issue, but it appears to improve the situation.

@massimozanetti are you using urllib3 version 1.x or 2.x?

@massimozanetti you provided a reproducer in https://github.com/SDFIdk/CDSETool/issues/67#issuecomment-1908043776, could you run that on your network with CDSETool 0.2.6 and see if the issue is still reproducible? When I upgraded to urllib3...

@massimozanetti not having randomness is a great step forward! Do you always get incomplete products, or only sometimes? Can you try a sequential version, so replacing: ```python #istantiate credentials credentials...

@massimozanetti Well, for better or worse, it's not the parallel downloads that are causing your issue. Can you try the sequential download but with urllib3 2.x, or is that difficult?...